Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when writing or correcting a lipid species name for a report or data table: render it in LIPID MAPS shorthand at the structural-resolution level the evidence supports — which cheaper models do not do by default.
.claude/skills/lipid-maps-shorthand-notation/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 7 |
| Model | Lift | Δ tokens | Δ turns | Cases | Verified |
|---|---|---|---|---|---|
| gemini-3.6-flashbest | +50% | +70% | 0% | 22 | 52d ago |
| gemini-3.5-flash | pending re-run | — | |||
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 122% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 84% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 156% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 71% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 5% | 0% |
Enforces the LIPID MAPS shorthand for mass-spectrometry-derived lipid names. The punctuation carries meaning: the separator between chains, the prefixes, and the hydroxyl suffix each encode exactly how much structure the measurement resolved. Apply whenever you write, normalize, or correct a lipid species name for a results table, figure, or report.
Class C:DB — the class abbreviation,ONE space, then carbons:double-bonds. Never fuse them (PC34:1) and never wrap the composition in parentheses (PC(34:1)). C is total acyl/radyl carbons, DB total carbon-carbon double bonds.
_./, sn-1 chain written first./ when a regiochemistry-resolving method was run (electron-activated dissociation,ozone-based, ultraviolet photodissociation, or an enzymatic assay). Ordinary collisional dissociation (CID) does NOT resolve sn — its acyl-loss intensities are a biased blend, not a structure readout — so default a CID-only name to _. When in doubt, drop a level.
Class C:DB (e.g. PC 34:1). Do not invent chains, an underscore, or a slash.
locant number followed by Z (cis) or E (trans), comma-separated for multiple bonds — 18:2(9Z,12Z). Never cis/trans words, 9c/9t, a Δ delta symbol, or an n-/omega descriptor.
O-,vinyl-ether plasmalogen = P- (e.g. PC O-16:0/18:1, PE P-16:0/20:4). The vinyl-ether double bond is implied by P- and is NOT counted in the chain's DB.
;O<n> on the sphingoid part:;O2 for a dihydroxy base (sphingosine/sphinganine), ;O3 for a trihydroxy base (phytosphingosine). The old d/t prefixes are deprecated — write Cer 18:1;O2/24:0, not Cer d18:1/24:0.
(LPC 18:1). Free fatty acids take the FA class token (FA 18:0), cholesteryl/sterol esters CE (CE 18:1). Added oxygens (hydroxyl/epoxide/keto) of unknown position are the same ;O<n> suffix on the affected part (FA 18:1;O = one extra oxygen), consistent with the sphingoid rule.
The same six punctuation rules apply across every lipid category — glycerophospholipids, glycerolipids, sphingolipids, sterols, and fatty acyls — so a name is portable between tools once it conforms.
PC(34:1) (parenthesized export) → PC 34:1PC 16:0/18:1 claimed from CID-only data → PC 16:0_18:1 (no sn evidence → underscore)PC 16:0/18:1 with EAD sn assignment → PC 16:0/18:1 (slash is licensed)Cer d18:1/24:0 (old dihydroxy prefix) → Cer 18:1;O2/24:0Cer t18:0/16:0 (old trihydroxy prefix) → Cer 18:0;O3/16:018:1 cis-9 / 18:1n-9 / 18:1Δ9 → FA 18:1(9Z)plasmenyl-PE 18:0/20:4 → PE P-18:0/20:4PC(O-16:0/18:1) → PC O-16:0/18:1LysoPC 18:1 / 1-oleoyl-LPC → LPC 18:1 (single acyl chain, no separator)cholesteryl oleate → CE 18:1 (one acyl chain on the sterol)18:1;OH (9-hydroxy, position unspecified) → FA 18:1;O (functional-group oxygen suffix)P-34:1 fromO-34:2 (a vinyl ether equals an ether plus one double bond). Report O- unless a vinyl-ether diagnostic ion or acid-lability test confirms P-.
P- double-bond bookkeeping: because P- implies the enol double bond, P-18:0 and O-18:1describe the same radyl; count only the additional double bonds under P-.
and omit the letter rather than guessing Z/E.
(e.g. _ to /) to look more precise.
¹³C-isotope artifact of an even neighbor; do not name one as an intact species without chain confirmation.
;O<n> count rather than asserting aspecific hydroxyl/epoxide locant the data did not localize.
_ when the sn assignment is unproven. DON'T emit / without a regiochemistry-resolving method.;O2 / ;O3. DON'T carry the old d / t sphingoid prefixes.(9Z) / (9E). DON'T write cis/trans, 9c/9t, Δ9, or n-9.O- and plasmalogen with P-. DON'T spell out "alkyl", "plasmenyl", or "plasmalogen"./) when only the chain composition is known — the single most common overstatement.d18:1 / t18:0 sphingoid prefixes instead of ;O2 / ;O3.cis/trans or with Δ / n- instead of (nZ) / (nE).PC(16:0/18:1)) or fusing the class to the numbers.C:DB (no parentheses)_ unless sn was truly resolved, then / with sn-1 first_//)(nZ) / (nE) for any double-bond position + geometryO- ether / P- plasmalogen prefix;O2 / ;O3 sphingoid hydroxyl, never d / t| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,596 | 6,500 | -58% | 1 | 1 | 0% | 1,600 | 3,553 | +122% | 0 | 0 | — |
case-02 | fail→pass | 8,148 | 4,515 | -45% | 1 | 1 | 0% | 1,657 | 3,053 | +84% | 0 | 0 | — |
case-03 | fail→fail | 9,884 | 4,957 | -50% | 1 | 1 | 0% | 2,208 | 3,153 | +43% | 0 | 0 | — |
case-04 | fail→pass | 5,154 | 2,213 | -57% | 1 | 1 | 0% | 933 | 2,392 | +156% | 0 | 0 | — |
case-05 | fail→pass | 8,093 | 2,963 | -63% | 1 | 1 | 0% | 1,513 | 2,582 | +71% | 0 | 0 | — |
case-06 | fail→pass | 13,292 | 3,484 | -74% | 1 | 1 | 0% | 2,510 | 2,627 | +5% | 0 | 0 | — |
case-07 | fail→fail | 5,290 | 3,034 | -43% | 1 | 1 | 0% | 1,037 | 2,504 | +141% | 0 | 0 | — |
case-08 | fail→pass | 8,490 | 2,808 | -67% | 1 | 1 | 0% | 1,611 | 2,565 | +59% | 0 | 0 | — |
case-09 | fail→fail | 5,029 | 2,752 | -45% | 1 | 1 | 0% | 912 | 2,551 | +180% | 0 | 0 | — |
case-10 | fail→pass | 5,503 | 2,152 | -61% | 1 | 1 | 0% | 1,084 | 2,432 | +124% | 0 | 0 | — |
case-11 | fail→fail | 11,814 | 3,238 | -73% | 1 | 1 | 0% | 2,108 | 2,584 | +23% | 0 | 0 | — |
case-12 | fail→fail | 4,655 | 3,121 | -33% | 1 | 1 | 0% | 869 | 2,587 | +198% | 0 | 0 | — |
case-13 | fail→fail | 7,147 | 3,716 | -48% | 1 | 1 | 0% | 1,362 | 2,707 | +99% | 0 | 0 | — |
case-14 | fail→fail | 8,277 | 12,178 | +47% | 1 | 1 | 0% | 1,822 | 4,445 | +144% | 0 | 0 | — |
case-15 | fail→pass | 10,274 | 5,195 | -49% | 1 | 1 | 0% | 1,821 | 2,991 | +64% | 0 | 0 | — |
case-16 | fail→pass | 23,006 | 3,039 | -87% | 1 | 1 | 0% | 1,843 | 2,579 | +40% | 0 | 0 | — |
case-17 | fail→pass | 11,025 | 2,816 | -74% | 1 | 1 | 0% | 1,929 | 2,547 | +32% | 0 | 0 | — |
case-18 | fail→pass | 13,110 | 4,143 | -68% | 1 | 1 | 0% | 2,302 | 2,824 | +23% | 0 | 0 | — |
case-19 | fail→fail | 22,073 | 6,409 | -71% | 1 | 1 | 0% | 3,618 | 3,149 | -13% | 0 | 0 | — |
case-20 | fail→fail | 7,474 | 2,881 | -61% | 1 | 1 | 0% | 1,395 | 2,650 | +90% | 0 | 0 | — |
case-21 | fail→fail | 8,425 | 4,849 | -42% | 1 | 1 | 0% | 1,321 | 2,925 | +121% | 0 | 0 | — |
case-22 | fail→fail | 8,957 | 5,575 | -38% | 1 | 1 | 0% | 1,764 | 3,189 | +81% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +50 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.5-flash | verified | 7/9/2026 | +96% |
Other measured skills in the registry, with their headline benchmark lift.